TY - GEN
T1 - Individual Functional Network Abnormalities Mapping via Graph Representation-Based Neural Architecture Search
AU - Li, Qing
AU - Dai, Haixing
AU - Lv, Jinglei
AU - Zhao, Lin
AU - Liu, Zhengliang
AU - Wu, Zihao
AU - Wu, Xia
AU - Coles, Claire
AU - Hu, Xiaoping
AU - Liu, Tianming
AU - Zhu, Dajiang
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
PY - 2023
Y1 - 2023
N2 - Prenatal alcohol exposure (PAE) has garnered increasing attention due to its detrimental effects on both neonates and expectant mothers. Recent research indicates that spatio-temporal functional brain networks (FBNs), derived from functional magnetic resonance imaging (fMRI), have the potential to reveal changes in PAE and Non-dysmorphic PAE (Non-Dys PAE) groups compared with healthy controls. However, current deep learning approaches for decomposing the FBNs are still limited to hand-crafted neural network architectures, which may not lead to optimal performance in identifying FBNs that better reveal differences between PAE and healthy controls. In this paper, we utilize a novel graph representation-based neural architecture search (GR-NAS) model to optimize the inner cell architecture of recurrent neural network (RNN) for decomposing the spatio-temporal FBNs and identifying the neuroimaging biomarkers of subtypes of PAE. Our optimized RNN cells with the GR-NAS model revealed that the functional activation decreased from healthy controls to Non-Dys PAE then to PAE groups. Our model provides a novel computational tool for the diagnosis of PAE, and uncovers the brain’s functional mechanism in PAE.
AB - Prenatal alcohol exposure (PAE) has garnered increasing attention due to its detrimental effects on both neonates and expectant mothers. Recent research indicates that spatio-temporal functional brain networks (FBNs), derived from functional magnetic resonance imaging (fMRI), have the potential to reveal changes in PAE and Non-dysmorphic PAE (Non-Dys PAE) groups compared with healthy controls. However, current deep learning approaches for decomposing the FBNs are still limited to hand-crafted neural network architectures, which may not lead to optimal performance in identifying FBNs that better reveal differences between PAE and healthy controls. In this paper, we utilize a novel graph representation-based neural architecture search (GR-NAS) model to optimize the inner cell architecture of recurrent neural network (RNN) for decomposing the spatio-temporal FBNs and identifying the neuroimaging biomarkers of subtypes of PAE. Our optimized RNN cells with the GR-NAS model revealed that the functional activation decreased from healthy controls to Non-Dys PAE then to PAE groups. Our model provides a novel computational tool for the diagnosis of PAE, and uncovers the brain’s functional mechanism in PAE.
KW - Brain Network Decomposition
KW - Graph Representation-based Neural Architecture Search
KW - Prenatal Alcohol Exposure
KW - fMRI
UR - http://www.scopus.com/inward/record.url?scp=85177428315&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-46671-7_6
DO - 10.1007/978-3-031-46671-7_6
M3 - Conference contribution
AN - SCOPUS:85177428315
SN - 9783031466700
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 79
EP - 91
BT - Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings
A2 - Yang, Xiaochun
A2 - Wang, Bin
A2 - Suhartanto, Heru
A2 - Wang, Guoren
A2 - Jiang, Jing
A2 - Li, Bing
A2 - Zhu, Huaijie
A2 - Cui, Ningning
PB - Springer Science and Business Media Deutschland GmbH
T2 - 19th International Conference on Advanced Data Mining and Applications, ADMA 2023
Y2 - 21 August 2023 through 23 August 2023
ER -